Several computational classifiers were developed using over 10000 screened compounds collected from NCGC against five major CYP450 isozymes of 1A2, 2C9, 2C19, 2D6, and 3A4. Random forest was used to develop models for these five isozymes using a set of MOE 2D descriptors. Five-fold cross-validation strategy was implemented to ensure the internal predictive ability of the models. The rigorously validated models exhibited outstanding predictive power for all five CYP450 isozymes with correct classification rates of 84.4%, 82.5%, 82.1%, 78.4%, and 80.0% for 1A2, 2C9, 2C19, 2D6, and 3A4, respectively. Structural analysis showed that halogenated (chlorine/fluorine) aromatics have significant higher frequency in CYP450 inhibitors than that in non-inhibitors. This comprehensive study yielded a compendium of validated QSAR models, which were then used to virtual screen a set of 954compounds in ToxCast project initiated by US EPA. Case studies confirmed the predictability of QSAR models through comparing the putative properties with the activities of several compounds. The proposal models carried with robustness and reliability and could be used to virtual screening other environmental chemicals for tentative risk assessment.